Identification Research Of Watermelon Leaf Diseases Based On Image Segmentation

Timely detection and prevention of crop leaf diseases are important to increase crop yield and to ensure food security.However,the limitation of farmers’ education level and the shortage of agricultural technical professionals greatly restrict detection and prevention.It is of great significance to develop automatic identification algorithms and software tools for leaf diseases based on computer vision technology.In recent years,deep learning techniques,especially convolutional neural network models,have achieved great success in the fields of computer vision and speech recognition.However,deep learning models require large-scale data sets for training,and the time limitations of data set construction and huge manpower and material resources restrict the application of deep learning models.Aim at the problem of complex image background and high cost of data set construction in the application scenarios of using mobile handheld devices to take pictures to identify watermelon leaf diseases,this paper carries out research on image segmentation and classification and recognition based on small-scale data sets,and develops a prototype application system.The specific work is as follows:(1)The process of segmenting the diseased leaves from the complex background and then identifying the leaf diseases through the classification model is proposed.Compared with the direct classification and recognition of the captured images,the recognition accuracy is increased by 4 percentage points.(2)Propose a deep neural network model which utilizes semantic segmentation of diseased leaves and extension of data generated through GAN training.The segmentation and intersection on the extended data set is increased by 2.3 percentage points compared with the general flipped extension method.(3)Propose a watermelon-infected leaf recognition model INC-VGG based on transfer learning.Compared with ordinary data augmentation,The value of F1 is increased by 1.7percentage points.(4)Apply the developed model to handheld devices,and develop a watermelon leaf disease identification system based on the android platform,which can basically meet the needs of agricultural practitioners.